Human data / Workflow demonstrations

Real tasks.
Human actions, captured.

Build training and evaluation data from people doing the work. We manage software and physical task demonstrations, capturing actions, narration, corrections, and outcomes in a structured dataset.

Scope your workflow data project

What we capture

The steps between intention and outcome.

A finished result tells only part of the story. Workflow demonstrations show the choices, intermediate steps, and corrections that led there, organized around the task your model needs to learn.

01

Software workflows

Capture people completing defined tasks in software, with screen recordings and action records appropriate to the environment. Preserve the sequence from starting state to final result.

02

Physical tasks

Record bounded, real-world tasks using a project-specific video protocol. Define the workspace, required views, task stages, and observable completion criteria before collection.

03

Decisions & recovery

Include participant narration where useful, and label corrections, retries, and outcomes. Document what happened when the work departed from the expected path.

Collection & quality assurance

A defined task.
A managed collection.

We turn your workflow into collection instructions, participant requirements, and review criteria. Every delivery is organized around the actions and outcomes your training or evaluation pipeline needs.

  1. 01

    Scope the workflow

    Agree on the task boundaries, starting conditions, capture method, success criteria, and output schema. Specify how to record incomplete or unsuccessful attempts.

  2. 02

    Prepare participants

    Recruit people suited to the task and qualify them against the collection protocol. Confirm the environment and recording setup before production.

  3. 03

    Capture & annotate

    Manage collection and review recording completeness. Annotate task stages, action sequences, corrections, and outcomes with references to the source timeline.

  4. 04

    Review & release

    Check labels against the recordings, resolve ambiguous cases, and verify structure and source references against the agreed delivery requirements.

What you receive

Recordings with
the work made legible.

Receive the source evidence and structured records together, so your team can connect each labeled action and outcome to the task that produced it.

Task recordings
Screen or video recordings, with participant narration and transcripts where included in the scope.
Action and task-stage records
Ordered actions and temporal annotations linked to source recordings, with task and attempt identifiers.
Outcome and correction labels
Completion status, retries, corrections, and project-defined outcome categories reviewed against the task criteria.
Structured trajectories
Sequences of task stages, actions, and results delivered in your required schema, with the collection protocol and label definitions.

Start with a workflow

What should your model
learn to do?

Share the task, the working environment, and the data your team needs. We will scope the participants, capture protocol, annotation workflow, and finished dataset.

Scope your workflow data project